Which AI meeting assistants actually speak in the meeting?

Most AI meeting assistants record and summarize. A few answer a spoken question. Very few take part. A sourced look, as of November 2026, at what each kind does in the call.

By Taylor Blake, Founder, Mainroom · Published November 1, 2026 · Updated November 1, 2026

"AI meeting assistant" covers very different products. The useful question is what the AI does while the meeting is happening. There are three levels.

1. Records and summarizes

Most products. A bot or a desktop app captures the audio and produces notes after the call. Examples, from each vendor's own pages as of November 1, 2026:

• Fathom records and summarizes meetings on Meet, Zoom and Teams, with a free plan. (fathom.ai)
• Granola is a notepad that writes notes "without a meeting bot" by recording device audio. (granola.ai)
• tl;dv records and summarizes, with a bot-free desktop option. (tldv.io)
• Fellow adds agendas and recaps to AI notes. (fellow.ai)

None of these speak in the call.

2. Answers a spoken question

A newer feature in notetakers: say a wake word and the bot answers a question about the meeting out loud.

• Fireflies documents "Hey Fireflies" voice answers in Google Meet, Teams and Zoom, on all plans. (Fireflies help)
• Otter announced "Hey Otter" voice answers for Zoom in March 2025, with Teams and Meet to follow. (announcement)

This is useful, and it is still a notetaker: it answers about the meeting, it does not take part in it.

3. Takes part

An AI participant that speaks when addressed and also acts in the meeting. The platform assistants act inside their own app: Microsoft Facilitator runs the agenda, timers and notes in Teams meetings, in chat and on the meeting stage (Microsoft support). A few independent products put a speaking agent into the call, including Mainroom.

A Mainroom agent joins Google Meet, Microsoft Teams or Zoom from a calendar invite and, during the call:

• answers when addressed, waits for a gap, and stops when a person talks over it;
• puts slides, tables and charts on the shared screen;
• teaches on a whiteboard it draws step by step as it explains;
• runs a live browser demo the room steers;
• looks things up in your tools (any MCP server) and the web;
• opens an anonymous intake for the room;
• coaches the person who invited it on a private companion page;
• logs decisions and commitments, then emails a recap.

Which level do you need?

If you only need a record, a notetaker is simpler. If you want quick answers about what was said, a wake-word notetaker is enough. If you want the meeting itself to go differently (questions answered in the call, material on the screen, input from the whole room), you need a participant.

Details about other products come from their public pages as of November 1, 2026 and may have changed. Corrections: hello@mainroom.ai.